Build a brand dashboard your team can keep using

Start with the accounts your team already uses

Open the brand’s Facebook Page, Instagram account, LinkedIn Page and Google Search Console. Sign into the correct accounts and check that you can view or export their insights. If an approved connector or native export is already available, the agent can use that instead.

In ChatGPT Work, connect the official browser extension through Settings → Computer Use, then select the signed-in browser or tabs. Computer Use is a fallback where enabled. In Claude, use its connected Chrome workflow where available. If access fails, the agent should identify the account and exact export or sign-in step needed; you handle passwords and two-factor authentication yourself.

Use four sources as the starting point

The main prompt chooses the last completed calendar month, compares it with the month before, and uses the sources below. You can change those choices later. Google Search stays separate from social activity; its clicks are not social referrals or leads.

Scroll across to see every column if needed ↔

Default monthly reporting sources

Source	Collect where available	Keep the meaning clear
Meta: Facebook & Instagram	Facebook Page and Instagram account: reach or native views/impressions, interactions, clicks and follower change where available.	Collect Facebook and Instagram separately. Keep native definitions and organic/paid scope.
LinkedIn Page Insights	Page impressions, native clicks, interactions and follower change; top content.	Use the Page’s own analytics. Campaign Manager is a separate paid source.
Google Search Console Insights	Search clicks, impressions, CTR and average position; leading pages/queries.	Use the requested calendar month and native totals. Search visibility is separate from social activity.

Keep the native metric names and reporting timezones. Use Search Console’s underlying Performance data when needed for exact calendar-month totals. Never replace a calendar month with a rolling-period summary.

Check the figures, then keep the commentary short

Retain the source exports and record the account, period, filters and extraction time. Keep each source’s totals and definitions intact: reach can overlap across platforms, rates have different denominators, and post breakdowns should not be added to account totals.

Show missing data as missing. A zero needs evidence, and a partial export should not become a complete-month result. Use absolute changes and percentage points where appropriate; avoid a percentage-change claim when its comparison base is zero or unavailable.

Give the team a visual brand console

The dashboard is the main communication tool: a clear latest-month view, a few highlights and a history the team can explore. The prototype below uses a fictional brand and dataset to show that idea. Change the month to move between the latest result and earlier performance.

Use the brand kit already supplied or confirmed in the agent’s context. If it is missing, ask for it once and continue with a neutral draft. Save approved colours, typography and asset references with the reporting brief so another teammate gets the same result.

Use ChatGPT Sites or a Claude artifact for the dashboard where available, with self-contained HTML as a fallback. In ChatGPT, @Visualize can help review an interactive draft inside the chat. Keep PDF and Excel as optional exports from the same data; they should support sharing rather than dictate the dashboard design.

Keep one source of truth for the history

A simple starting point is reporting/metrics.csv: one accepted monthly observation per account, metric and reporting period. Keep the raw exports, metric definitions and a correction log beside it. The dashboard reads that checked history each time.

On a rerun, match the existing record and replace it only with a validated correction; do not append the same month again. Preserve earlier months and record the old and new values. Stage partial or overlapping exports separately: daily reach, month-to-date figures and monthly totals cannot simply be added together. A failed refresh should flag the gap while preserving previously accepted history.